This may seem obvious to some, but it might not be an obvious consideration for others. Who would even normally think a really old socket 939 CPU fan/heatsink assembly
would fit on an AM3 socket CPU
? Hence, the sharing of info.
I saved money (always nice), and still
got improved cooling with "old" STOCK fan/heatsink assemblies.
I also got a noticeably much quieter system when running at typical CPU loads.
Note that the same "up-front" investigation of what you have on hand might let you buy an "OEM" CPU and thus get a higher performing CPU for the same money you would have spent for a lower-performing CPU that came with a stock heatsink assembly, or if you had to buy a separate cooling assembly.
This all came about because I just got a CPU that "needed" a tiny bit more cooling. I couldn't see buying and installing/hassling-with an entire new >$30 heatsink assembly just to bring the temp down about 10C, so I started checking some of my CPU fans from OLD systems and made some quick discoveries.
First I looked at simply replacing the "new" fan with an "old" fan, then I looked at changing to use the "old" heatsinks, too. Changing-out JUST the fan is not only less risk, but this yields a direct fan-to-fan comparison. No changes introduced in mating the heatsink to the CPU, thermal compounds, alignment, etc. The only
variable was the fan.
"OLD" CPU Fan Findings
I had first tried the simplest change (the fan) because that can be done without even removing the assembly from the motherboard (in this particular heatsink/fan design the fan is affixed to the heatsink with just the 4 screws).
THEN I LOOKED AT OLD HEATSINKS
I found some of the "old" fans have significantly higher RPM capability than the fans being shipped on stock CPU heatsinks these days...as much as 100% higher RPM, while needing only about 20% more milli-amps (e.g. 400mA of "old" fan v. 330mA of "new" fan) at max RPM.
I was at first VERY surprised, but thinking back on things, I shouldn't have been. Due to CPU fabrication processes, the older fan/heatsink assemblies were designed for CPUs that ran at the same or even higher Watts than today's CPUs. For example, the socket 939 Athlon64 X2 4600+ ran at about 110Watts...for 2 cores...and the "stock" fan/heatsink for that CPU was designed for that high power output. That "old" assembly could easily handle any of today's 95W CPUs, and would likely be an improvement. Many of the 939 socket CPUs seem to have run at about 90Watts, but some later versions ran at 67Watts (due to the manufacturing process change).
For example, off an old socket 939 AMD Athlon64 3800+ I found a 70mm, dual ball-bearing fan that runs at 5,500rpm @12volts. That's nearly TWICE the speed of any stock CPU fan that's come with any AMD CPU (1,2,3, or 4 core) recent purchases. As a plus, the amount of noise is almost identical at full speed, even though the ft3/sec--m3/sec air volume output is easily determined to be higher...much much higher with the "old" 939 fan. This reduced noise with higher output is likely due to bearing and blade design differences. At 2,000rpm, the "old" fan was much quieter than the "new" fan. I chose 2,000rpm as the comparison-point as that is about where the "new" CPU fan normally ran during typical low-demand usage on the test system. Also, the fans were removed from their respective heatsinks for these comparisons.
The stock CPU AMD fan that came with my new CPU maxes out at ~2,500rpm@12Volts, and was letting temps reach 65C (and still climbing) within moments of running Prime95 on all cores as close to 100% as I can get it. But, the OLD 939 fan under the same conditions kept the temp stable and below 58, and while getting there the temps increased much more slowly. It took minutes instead of seconds to get to 58C (note: this is explained by the fan controller increases fan rpm in X% increments...obviously X% of a higher speed fan is higher than X% of a lower speed fan. But, for each X% increase in rpm of the "new" fan you get more than that X% additional air flow/cooling you get with the "old" fan, which means it takes longer to heat up for a given X% increase in rpm with the "new" fan)
But after that success with the "old" fan, I decided to check the OLD heatsinks I have and found that several 939 socket heatsinks mated perfectly with the AM3 socket. If needed, you can even exchange the cam "rocker-bar" assembly in some cases while some cam "bars" didn't even need to be changed. And in this case all relative vertical mounting-related dimensions were maintained with the cam both open and closed. With this AM3 socket design, this means the forces applied to the top of the CPU and the socket/mobo are the same. No added stresses=Totally interchangeable. Still...exercise EXTRAORDINARY caution every step of the way when trying an interchange of heatsinks and/or cam "bars"
Using that old 939 heatsink, the current(new) cam assembly, and the old 3800+ CPU fan, I got another 5C reduction. (see notes 1-3)
1) On the heatsink comparisons, since my number of tests is so low, it is possible the differences were due to the way I assembled the unit.
2) The "old" heatsinks almost all added at least 1cm to the height of the assembly, and provided more surface area.
3) I installed the "old" heatsinks using what I had on hand
: some ~12 month old factory sealed CoolerMaster's "Ice Fusion". A definitely non-premium thermal compound. I'm very skilled at applying thermal compound on hundreds of electronic heatsink designs, but even so I would bet the cooling improvement would have been substantially better had I used a decent Arctic Silver compound. I'll probably try that later, but since my system cooling is working fine as is, I'll probably wait until I see another system/CPU.
edit: minor typos, a couple clarifications, a name correction